Multi-scale target detection model method based on metric learning

Abstract

The invention discloses a multi-scale target detection model method based on metric learning, and the method comprises the steps: carrying out the initialization of a network through employing a pre-trained model through employing the idea of transfer learning, adding a loss function to carry out the fine adjustment of a weight parameter of the network, and improving the regression precision of aregion of interest in an image; adding a full-connection layer to vectorize feature information after a feature map is extracted from a region of interest in an image, and then performing distance measurement by using the feature information extracted by the full-connection layer to realize classification and identification of target region information. By adopting the method of the invention, theloss of the feature information can be reduced, and the classification and recognition accuracy of the detection target is improved.

Description

technical field [0001] The invention relates to image processing and small-sample target detection technology, in particular to a multi-scale target detection model method based on metric learning, which belongs to the technical fields of computer vision and edge computing. Background technique [0002] In recent years, deep learning has achieved great success in image classification and detection tasks. Because of its superior performance in image feature extraction, it can learn the feature information required for tasks from data and is widely used in Various areas of research in computer vision. For the existing deep neural network, due to its complex network structure with a large number of parameters, a large amount of data is required for training to complete specific computer vision tasks. However, in some practical applications, it is very difficult to collect a large amount of data, so how to use the existing small amount of data to achieve specific computer visio...

AI Technical Summary

Problems solved by technology

For the existing RepMet network model, solve the following two problems: (1) How to perform the regression frame regression of the region of interest in the image when using a small number of samples for model training; (2) Identify the region of interest in the image How to reduce the loss of feature information and improve the final classification and recognition accuracy

Images